Write CCTV analytics software requirements that vendors can be tested against. Define scenarios, false alarm limits, licensing and on site acceptance testing.
Many tenders ask for CCTV analytics software with a single line such as "AI analytics required". Every vendor then ticks the box, and the buyer only discovers the gaps after handover. False alarms flood the control room, or intrusions pass unnoticed at night. This guide shows facility and procurement teams how to turn analytics into measurable requirements. You will learn which analytics to specify, how to define accuracy and false alarm limits, what to ask about licensing and how to run acceptance tests on site.
Key Takeaways
Specify camera analytics by scenario, distance and lighting, not by feature names in a brochure.
Set a false alarm limit per camera per day and an agreed tuning period before final acceptance.
Make payment milestones depend on an on site acceptance test of your cctv video analytics, not on a vendor demo.
Types of Video Analytics in CCTV Worth Specifying
Start by listing only the analytics your site actually needs. Each function adds licences, processing load and tuning effort, so a shorter list tested properly beats a long list nobody uses.
The common types of video analytics in CCTV fall into a few groups. Perimeter analytics cover line crossing and intrusion zones. Behaviour analytics cover loitering, running and crowd density. Object analytics cover items left behind or removed. Identification analytics cover number plates and faces. Business analytics cover people counting and queue length. For background on how these tools are shaping security, see our article on video analytics in smart cities.
Map each analytics type to a real risk area before you write a single requirement. A warehouse may need intrusion detection on the fence, loitering alerts at the loading bay after hours and plate reading at the gate. It rarely needs crowd density or queue analytics. A shopping mall is the opposite. Writing this map into the tender stops bidders from padding proposals with features that add licence costs but deliver no value. It also tells bidders which cameras carry analytics load, so they can size servers and choose suitable camera models from the start.
Be clear about what is out of scope too. If you only need alerts during closed hours, say so. Schedules reduce false alarms and processing load, and they make acceptance tests easier to plan.
Writing Measurable Analytics Requirements
A requirement is only useful if you can test it on site. Describe the scenario, the conditions and the pass mark for every analytics rule you include.
For each rule, state the target, the distance, the lighting and the expected result. For example: "Detect a person crossing the north fence line at up to 40 metres, day and night, with at least 95 percent of test crossings detected." Add the conditions that must not trigger alerts, such as swaying palms, birds, headlights and shadows. Then set a false alarm limit, such as a maximum number of nuisance alerts per camera per day after tuning. Agree these figures with your security consultant, because realistic targets depend on the site.
Avoid vague words such as "accurate", "intelligent" or "minimal false alarms". They cannot be tested, so they cannot be enforced. Replace them with numbers, conditions and a method of measurement. If a vendor claims a high accuracy figure, ask under what conditions it was measured and whether the same test can be repeated on your site. Independent test libraries such as the UK Home Office imagery library for intelligent detection systems show how structured evaluation scenarios are built, which is a useful model for your own script.
Require the CCTV analytics software to log every alert with the rule name, camera, time and a short clip. This log becomes the evidence base for acceptance testing and for later performance reviews under maintenance.
Name the evidence you expect too. Ask for alarm clips, snapshots and time stamps that operators can export. Where face or plate recognition is included, ask vendors which independent tests their algorithm has taken part in, such as the NIST face recognition vendor tests, and check local privacy rules before enabling it.
Edge, Server and Licensing Choices
Where analytics run affects cost, bandwidth and resilience. Your tender should state the preferred architecture or ask bidders to justify their choice.
Edge analytics run inside each video analytics camera. They reduce network load and keep working if the server fails, but capabilities vary by camera model. Server analytics run on central hardware and can apply advanced rules to many camera brands, but they need GPU capacity and strong network links. Many sites combine both. Ask bidders to state the licence model for every rule, whether licences are perpetual or annual, and what happens when you add cameras.
Require open integration as well. Analytics events and metadata should reach your video management system, access control and alarm platforms. Supporting ONVIF Profile M for analytics metadata reduces lock in when you change or add brands. Our CCTV design and installation team can review integration clauses before you issue the tender.
Hardware sizing belongs in the tender as well. Ask each bidder to state how many analytics streams each server supports, the GPU model proposed and the spare capacity left for growth. Without these figures, CCTV analytics software that performs well on ten cameras can slow down badly at fifty. Ask for the upgrade path and its cost, so future expansion does not become a surprise budget request.
Acceptance Testing and the Tuning Period
Acceptance testing proves the system meets your written requirements under real conditions. Plan it in the tender so bidders price the time and effort.
Write a test script for each camera and rule. Include walk tests at set distances, tests at night and in bright glare, and tests with the nuisance conditions you listed. Record every result in a signed log. After the first test, allow a tuning period, often a few weeks, where the installer adjusts zones, sensitivity and schedules while you track false alarms per camera per day. Final acceptance follows only when detection and false alarm targets are met. Link ongoing performance checks to your annual maintenance contract so accuracy does not drift when layouts, lighting or seasons change.
Retain a portion of the contract value until final acceptance is signed. This keeps the installer focused on tuning rather than moving to the next project. Also require a handover pack with the final rule settings, zone drawings, schedules and test logs, so a future maintainer can restore the configuration after a hardware replacement.
What to Ask Bidders to Submit
A structured submission makes proposals comparable. Ask every bidder to answer the same questions in the same format, so price differences reflect real scope differences.
A compliance table showing how each analytics requirement will be met, camera by camera.
The CCTV analytics software name, version, licence model and renewal costs for five years.
Server and GPU specifications with the number of analytics streams supported.
Integration details for your video management, access control and alarm systems.
A draft acceptance test script and the proposed tuning period.
References from similar sites where the same analytics are running today.
Score bidders on evidence, not claims. A bidder who offers a site demonstration on your own cameras gives you far more confidence than a polished video. For a view of how newer camera hardware supports these requirements, read about AI security cameras in Dubai.
Conclusion
Clear requirements turn CCTV analytics software from a marketing promise into a measurable tool. Specify only the analytics you need, define scenarios and false alarm limits, choose an architecture that fits your network and prove results through on site acceptance testing. Better hardware makes these targets more achievable, but only when the tender demands evidence. Our CCTV maintenance and support team can then keep performance on track year after year.
Planning a security tender? Talk to Vega Digital's design engineers for help writing analytics requirements and test scripts.
Frequently Asked Questions
What is CCTV analytics software?
It is software that analyses live or recorded video to detect events automatically, such as intrusion, loitering, abandoned objects or number plates, and sends alerts to operators instead of relying on people watching every screen.
What types of video analytics are used in CCTV?
Common types include line crossing, intrusion zones, loitering, object left or removed, crowd density, people counting, number plate recognition and face recognition. Most sites need only a few of these, chosen by risk area.
How do I measure analytics accuracy in a tender?
Define test scenarios with distances, lighting and targets, then set a minimum detection rate and a maximum false alarm rate per camera per day. Verify both through signed on site tests before final acceptance and payment.
What is an acceptable false alarm rate?
There is no universal figure. The limit depends on site size, environment and how many operators respond. Agree a realistic number per camera per day with your consultant, then enforce it after a defined tuning period.
Should analytics run on the camera or the server?
Edge analytics suit simple rules and reduce bandwidth, while server analytics handle complex rules across many brands. Many projects combine both, so ask bidders to justify their architecture and show the resulting network and hardware requirements.
Does camera analytics need a strong network?
Server based analytics need reliable bandwidth to stream video from every camera. Our business network installation team designs switching and links so analytics and recording both run smoothly without dropped frames or delayed alerts.
How long does analytics tuning take?
Tuning often takes a few weeks after the first test, because weather, lighting and daily activity vary. Track false alarms per camera during this period and adjust zones and sensitivity before final acceptance is signed.
Can analytics software work with any camera brand?
Many server platforms support multiple brands, especially cameras following ONVIF profiles. Edge analytics usually depend on the camera maker. Confirm compatibility for every proposed camera model in writing within the tender before accepting any proposal.
Who should run the acceptance test?
The installer runs the tests while the client, consultant or facility manager witnesses and signs each result. Choose an installer with its own in house engineers so the same team designs, tests and supports the system.
